Distributed transaction processing system having resource managers that collaborate to decide whether to commit or abort a transaction in response to failure of a transaction manager

• US 8,650,272 B2
• Filed: 09/26/2008
• Issued: 02/11/2014
• Est. Priority Date: 07/16/2008
• Status: Active Grant

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Myers, Douglas B.
• Primary Examiner(s): Ibrahim, Mohamed
• Application Number:
  US12/238,779
• Publication Number:
  US 20100017642A1
• Time in Patent Office: 1,964 Days
• Field of Search: 709223-226, 718/104
• US Class Current: 709/223
• CPC Class Codes: G06F 11/0709 in a distributed system consisting of a plurality of standalone computer nodes, e.g. clusters, client-server systems
• CPC Class Codes: G06F 11/0757 by exceeding a time limit, i.e. time-out, e.g. watchdogs
• CPC Class Codes: G06F 11/1474 in transactions G06F16/20 takes precedence

Abstract

A distributed transaction processing system includes a plurality of resources, resource managers to manage corresponding ones of the resources, and a transaction manager to coordinate performance of a transaction with the resource managers. In response to failure of the transaction manager, the resource managers are configured to collaborate to decide whether to commit or abort the transaction.

Claims

1. A distributed transaction processing system comprising:

a plurality of resources;

resource managers to manage corresponding ones of the resources;

a transaction manager to coordinate performance of a transaction with the resource managers,

wherein in response to failure of the transaction manager, the resource managers are configured to collaborate to decide whether to commit or abort the transaction, wherein the collaborating comprises at least two of the resource managers communicating with each other, where the communicating includes a first of the at least two resource managers sending a first request to a second of the at least two resource managers to obtain a state of the transaction at the second resource manager, and the second resource manager sending a second request to the first resource manager to obtain a state of the transaction at the first resource manager; and

one or more processors, wherein the resource managers and transaction manager are executable on the one or more processors.

2. The distributed transaction processing system of claim 1, wherein, in response to failure of the transaction manager, each of the resource managers is to:

handle recovery and completion of the transaction with respect to the corresponding resource that the resource manager is managing.

3. The distributed transaction processing system of claim 2, wherein the first resource manager that has completed preparing the transaction but has not started either committing or aborting the transaction, is configured to:

commit or abort the transaction at the first resource manager according to the determined state of the transaction at the second resource manager obtained in response to the first request.

4. The distributed transaction processing system of claim 3, wherein the first resource manager is configured to:

commit the transaction in response to determining that the second resource manager has started to commit or has committed the transaction.

5. The distributed transaction processing system of claim 3, wherein the first resource manager is configured to:

abort the transaction in response to determining that the second resource manager (1) has started to abort or aborted the transaction;

or (2) has not finished preparing the transaction.

6. The distributed transaction processing system of claim 3, wherein the first resource manager is configured to:

abort the transaction in response to determining that all the other resource managers are in a prepared state for the transaction.

7. The distributed transaction processing system of claim 2, wherein each resource manager is configured to:

abort the transaction if the resource manager has not yet finished preparing the transaction for commit;

finish committing the transaction if the resource manager has already started committing the transaction;

finish aborting the transaction if the resource manager has already started aborting the transaction.

8. The distributed transaction processing system of claim 1, further comprising a persistent storage to store persistent information of the transaction for each of the resource managers.

9. The distributed transaction processing system of claim 8, wherein a particular one of the resource managers is configured to delete the respective persistent information of the transaction in response to determining that all of the other resource managers have already committed or aborted the transaction, and therefore will no longer require the corresponding persistent information for recovery purposes.

10. The distributed transaction processing system of claim 1, wherein the resource managers detect failure of the transaction manager based on a timeout condition, wherein a recovery process is to be started in response to the timeout condition, the recovery process comprising the resource managers collaborating to decide whether to commit or abort the transaction.

11. The distributed transaction processing system of claim 1, wherein failure of the transaction manager comprises either (1) the transaction manager crashing or (2) loss of communication with the transaction manager.

12. The distributed transaction processing system of claim 1, wherein the transaction manager is to communicate identities of the resource managers involved in the transaction to each of the resource managers involved in the transaction, wherein the collaborating uses the identities.

13. A method for use in a distributed transaction processing system, comprising:

providing resource managers to manage respective resources of the distributed transaction processing system;

providing a transaction manager to coordinate performance of a transaction with the resource managers;

configuring the resource managers with respective timeout conditions to cause the resource managers to timeout in response to failure of the transaction manager or any of the resource managers involved in the transaction; and

collaborating, by the resource managers involved in the transaction that are executing on one or more processors, to complete the transaction based upon timeouts of the resource managers, wherein the collaborating comprises at least two of the resource managers sending requests to each other regarding a state of the transaction.

14. The method of claim 13, wherein a particular one of the plural resource managers has completed preparing the transaction but has not started either committing or aborting the transaction, the method further comprising:

receiving a state of the transaction for at least one other of the resource managers in response to the request sent by the particular resource manager;

the particular resource manager committing or aborting the transaction according to the state of the transaction of the at least one other resource manager.

15. The method of claim 14, further comprising:

the particular resource manager committing the transaction in response to determining that the at least one other resource manager has started to commit or has committed the transaction; and

the particular resource manager aborting the transaction in response to determining that the at least one other resource manager (1) has started to abort or aborted the transaction;

or (2) has not finished preparing the transaction.

16. The method of claim 13, further comprising the transaction manager communicating the timeout conditions to the corresponding resource managers.

17. An article comprising at least one non-transitory computer-readable storage medium containing instructions that when executed cause a system having a processor to:

run resource managers to manage respective resources;

coordinate, by a transaction manager, performance of a transaction that involves plural resource managers;

detect failure of the transaction manager; and

in response to detecting failure of the transaction manager, send requests by the plural resource managers to each other to obtain states of the transaction at the corresponding resource managers to complete the transaction.

18. The article of claim 17, wherein a particular one of the lbural resource managers has completed preparing the transaction but has not started either committing or aborting the transaction, wherein the instructions when executed cause the system to:

receive, by the particular resource manager, a state of the transaction for at least one other of the plural resource managers in response to the request sent by the particular resource manager; and

commit or abort the transaction at the particular resource manager according to the state of the transaction for the at least one other resource manager.

19. The article of claim 18, wherein the instructions when executed cause the system to further:

commit, at the particular resource manager, the transaction in response to determining that the at least one other resource manager has started to commit or has committed the transaction.

20. The article of claim 18, wherein the instructions when executed cause the system to further:

abort, at the particular resource manager, the transaction in response to determining that the at least one other resource manager (1) has started to abort or aborted the transaction;

or (2) has not finished preparing the transaction.